1.
We read with great interest the multicenter pivotal study by Abe et al., the recent large‐scale real‐world study by Goto et al., and Suzuki's accompanying commentary on artificial intelligence (AI)‐assisted detection during esophagogastroduodenoscopy (EGD) [1, 2, 3]. Abe et al.'s retrospective video‐based study showed high detection rates with the CAD‐EYE prototype (FUJIFILM Corporation, Tokyo, Japan) for gastric neoplasms measuring ≤ 10 mm, whereas Goto et al. found an increased detection of gastric cancers of this size with CAD‐EYE‐assisted screening EGD [1, 2]. Such performance is important because AI‐assisted endoscopic detection aims to reduce endoscopist‐dependent variation and standardize cancer detection during EGD.
At a private community hospital, we previously demonstrated marked inter‐endoscopist variation in the detection of early gastric cancers measuring ≤ 10 mm during routine EGD, suggesting that these small lesions may be missed in daily clinical practice [4]. In an unpublished additional analysis of the same institutional endoscopic resection database, 99 of 278 early gastric adenocarcinomas treated by endoscopic resection had undergone EGD within 3 years before diagnosis. Thirty of the 278 lesions (10.8%) were classified as true missed cancers because they were clearly identifiable on prior EGD images, consistent with human recognition error. Their mean diameter was 12.9 mm (range, 1–38 mm).
Although Abe et al. [1] did not specifically evaluate true missed gastric cancers, the size‐stratified detection results for the CAD‐EYE prototype are relevant to our true missed cancer data. In our retrospective review, a representative true missed cancer was visible on archived prior white‐light images and was retrospectively highlighted by the gastroAI‐model G prototype (AI Medical Service Inc., Tokyo, Japan) (Figure 1) [5]. Viewed together, these findings suggest that true missed gastric cancers attributable to human recognition error may include lesions potentially detectable with AI assistance.
FIGURE 1.
Representative case of a true missed gastric cancer identified by retrospective review of prior EGD. The original endoscopic images used for all four panels were obtained during the same prior EGD; the white light images used for panels C and D were captured earlier than the narrow‐band and indigo carmine images shown in panels A and B. During this examination, a gastric cancer on the posterior wall of the antrum was recognized, whereas a second lesion at the lesser curvature of the antrum was not recognized, possibly because attention was directed toward the posterior wall cancer. Retrospective review later showed that the lesser curvature lesion was identifiable in panels A and B (white arrowheads). The lesion was subsequently detected on follow‐up endoscopy 1.5 years later and confirmed as gastric cancer after endoscopic resection. Panels (C, D) show the results of retrospective analysis of archived consecutive white light images using the gastroAI‐model G prototype. In panel (C), the posterior wall cancer was highlighted by the AI system. In panel (D), the visible but unrecognized lesser curvature lesion (white arrowheads) was highlighted by the AI system with a white rectangular frame.
In summary, AI‐assisted detection may help identify visible but missed gastric cancers attributable to human recognition error during routine EGD.
Author Contributions
D.M. contributed to the study conception and design, data collection, data interpretation and drafting of the manuscript. D.M. and M.Y. revised the manuscript in response to the editorial and reviewer comments. T.T. contributed to the artificial intelligence‐based image analysis and critically reviewed the manuscript for important intellectual content. M.Y. supervised the manuscript preparation and critically reviewed the manuscript for important intellectual content. All authors reviewed and approved the final version of the manuscript.
Funding
The authors have nothing to report.
Conflicts of Interest
Tomohiro Tada is the founder, Chief Executive Officer, and a shareholder of AI Medical Service Inc. The artificial intelligence‐based image analysis was performed using a system provided by AI Medical Service Inc. without financial compensation. The other authors declare no conflicts of interest.
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.
References
- 1. Abe S., Kitagawa Y., Hatta W., et al., “A Multicenter Pivotal Study on the Artificial Intelligence System for Neoplastic Lesions Detection in Upper Gastrointestinal Endoscopy,” Digestive Endoscopy 37 (2025): 1306–1314. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Goto C., Nakagawa R., Horio R., et al., “Impact of Artificial Intelligence‐Assisted Endoscopy on Screening for Upper Gastrointestinal Cancer in a Large‐Scale Health Screening Facility,” Digestive Endoscopy 38 (2026): e70159. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Suzuki S., “How Does Artificial Intelligence‐Assisted Endoscopy Impact Upper Gastrointestinal Screening?,” Digestive Endoscopy 38 (2026): e70215, 10.1111/den.70215. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Murakami D., Yamato M., Amano Y., Nishino T., and Arai M., “Variation in the Rate of Detection of Minute and Small Early Gastric Cancers at Diagnostic Endoscopy May Reflect the Performance of Individual Endoscopists,” BMJ Open Gastroenterology 10 (2023): e001143. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Hirasawa T., Aoyama K., Tanimoto T., et al., “Application of Artificial Intelligence Using a Convolutional Neural Network for Detecting Gastric Cancer in Endoscopic Images,” Gastric Cancer 21 (2018): 653–660. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.